Traveling Trends Social butterflies or frequent fliers?cosn.acm.org/2013/files/Session8/Session8Paper1.pdf · Traveling Trends Social butterflies or frequent fliers? with O. Varol,

Traveling Trends Social butterflies or frequent fliers?

with O. Varol, F. Menczer, A. Flammini

Emilio Ferrara - @jabawack

Center for Complex Networks and Systems Research

School of Informatics and Computing

Indiana University Bloomington (USA)

Trends and collective attention

October 8, 2013 ACM COSN 2013 – Emilio Ferrara – CnetS – Indiana University - @jabawack

What is a trend? – Baby names trends



What is a trend? – House pricing trends

The collaborative image of the city: mapping the inequality of urban perception

P Salesses, K Schechtner, CA Hidalgo. PloS one 8 (7), e68400, 2013

http://scholar.google.it/citations?view_op=view_citation&hl=en&user=xhCWdtMAAAAJ&sortby=pubdate&citation_for_view=xhCWdtMAAAAJ:k_IJM867U9cC



What is a trend? – Box office predictions

Predicting the future with social media. S Asur, BA Huberman. WI-IAT, 2010

Early p

redictio

n o

f movie

box o

ffice su

ccess b

ased o

n W

ikip

edia activity b

ig data

M M

estyán

, T Y

asseri, J K

erté

sz. PLoS O

NE 8

(8): e

71226, 2

013

http://scholar.google.it/citations?view_op=view_citation&hl=en&user=C_z8eaoAAAAJ&citation_for_view=C_z8eaoAAAAJ:d1gkVwhDpl0C

http://scholar.google.it/citations?view_op=view_citation&hl=en&user=AiLoMKwAAAAJ&sortby=pubdate&citation_for_view=AiLoMKwAAAAJ:8k81kl-MbHgC

What is a trend? – Financial market



Quantifying trading behavior in financial markets using Google Trends. T Preis, HS Moat, HE Stanley. Scientific reports 3, 2013

http://scholar.google.it/citations?view_op=view_citation&hl=en&user=lbJzOmEAAAAJ&sortby=pubdate&citation_for_view=lbJzOmEAAAAJ:7PzlFSSx8tAC

Gaming collective attention

October 8, 2013

How to game the system to drive collective attention

Persuasion: orchestrated manipulation, rumors, advertisement…

ACM COSN 2013 – Emilio Ferrara – CnetS – Indiana University - @jabawack

Toward detecting persuasion campaigns in social media. E Ferrara, O Varol, S Malinchik, F Menczer, A Flammini. Arxiv, 2013

Twitter trends dataset

October 8, 2013

Datasource:

We collect 63 US cities trends and also US national trends

Real-time monitoring (10 min intervals, 100% uptime)

Period: 50 days, from April, 12th 2013 to May, 31st 2013

We remove promoted hashtags

Total trends: 11,402

4,513 hashtags – 6,889 phrases


Results outline

Geo-temporal trend analysis

Trendsetters and trend-followers

A conjecture: Social butterflies or frequent fliers?


Trends spatial bimodal distribution:

Most trends are popular only in one/few cities

Many trends spread in all country

The remainder fail to achieve global popularity

Trends temporal distribution (lifetime):

Lifetime broadly distributed:

68% < 20m – 95% < 6h

0.3% > 1d

Entropy defined as

Trends reaching more places live longer

Low entropy: low expected lifetime

Spatio-temporal trend analysis


the time topic j trended in location i

Spatial trend similarity analysis


Shared trends

similarity:

G: South West

Y: Midwest

R: East Coast

Purple: ?

Hint: big cities!

Locality effect

Geography of trends


Geographic representation of the 63 cities and their respective clusters.

Clustering significance verification:

Compute distribution of similarity values

for all pairs of locations belonging to the

same cluster (intra-cluster similarity)

Compute distribution for all pairs

belonging to different clusters (inter-

cluster similarity)

Kernel smoothing:

Apply Kernel Density Estimation to

estimate the probability density

functions of the distributions

Significance:

T-test for any pair of distributions of

intra- and inter-cluster similarity to

determine if they might origin from the

same distribution

P < 0.01: the four clusters are

significant at the 99% C.L.

Significance of trends spatial clusters


Green Yellow Red Purple

Long Beach Memphis St. Luis Greensboro Washington Dallas-Ft.

Worth

Fresno Salt Lake City San Antonio Nashville New York Chicago

Mesa Harrisburg Milwaukee Norfolk Detroit Philadelphia

Tucson New Orleans Tampa Providence Boston Miami

Albuquerque Baton Rouge Pittsburgh Denver San Francisco Atlanta

Virginia Beach Portland New Haven Richmond Cleveland Los Angeles

San Jose Tallahassee Seattle Phoenix Minneapolis Raleigh

Colorado Springs San Diego Cincinnati Sacramento Las Vegas Houston

Jackson Kansas City Austin Columbus

Honolulu Oklahoma City Orlando Indianapolis

El Paso Birmingham Baltimore

Omaha Louisville

Jacksonville

Trends pathway analysis


Backbone extraction: α=0.3 (to keep only significant links) Directed edge bundling

Country backbone: East-to-West West-to-East

trendsetter trend-follower

Resutls outline







Location Rank ω(n)

Los Angeles 1st 0.806

Cincinnati 2nd 0.736

Washington 3rd 0.718

Seattle 4th 0.711

New York 5th 0.669

…

…

Kansas City 59th 0.352

Omaha 60th 0.352

El Paso 61st 0.235

Albuquerque 62nd 0.109

Oklahoma City 63rd 0.101

Q: Are trending topics that become popular

at the country level produced uniformly by

all cities, or preferentially by some of them?

Sources: trendsetters

Sinks: trend-followers

Weighted sink-source ratio:

Top ranked cities: 4/5 major metro areas

Los Angels & NY: also top worldwide HT

producers (Kamath et al. WWW 2013)

All sinks happen to be in the Midwest or

Southwest of the country

National trendsetters and trend-followers


Trendsetters vs. trend-followers

X: no. times a topic trending in a given city later becomes a national trend; Y: the reverse effect

Inset: a Gaussian Mixture Model identifying two different trendsetting dynamics

Results outline





Social butterflies or frequent fliers?

October 8, 2013

Q1: Does mere city size explain the trendsetting dynamics?

Larger cities produce more tweets; this yields to more potential topics competing for collective

attention, but the maximum number of possible trends is fixed to 10 at the same time!

As a result, the effect of sheer volume is discounted by construction in the definition of Twitter trends

Q2: Why the metro areas play such a trendsetting role, then?


16/17 purple cities are also top 20 air

traffic hubs!

Major travel cities including Atlanta,

Chicago, Los Angeles

Some purple cities are not in top 30

most US populated metro areas…

Charlotte, Raleigh, and Las Vegas appear

among the major traffic hubs!

Does information travel faster by

airplane than over the Internet?

Location Rank Traffic Class

New York 1st 54M |||||||||

Atlanta 2nd 45M |||||||||

Chicago 3rd 41M |||||||||

Miami 4th 33M |||||||||

Dallas-Ft. Worth 5th 32M |||||||||

Washington 6th 31M |||||||||

Los Angeles 7th 31M |||||||||

Denver 8th 25M |||||||||

Charlotte/Raleigh 9th 24M |||||||||

Houston 10th 24M |||||||||

San Francisco 11th 21M |||||||||

Las Vegas 12th 20M |||||||||

Conclusions and future work

October 8, 2013

Trends reflect a locality effect: they diffuse locally more

than globally in three specific geographical areas

East-coast, Midwest, South West

There is a fourth class of metropolitan cities

They are spread all over the country

They act as trendsetters (they produce most national trends)

They correspond to major air traffic hubs

Open questions:

What’s the role of traffic hubs in trend diffusion?

What’s the role of people?!


Thanks! Questions? @jabawack

Emilio Ferrara Onur Varol Filippo Menczer Alessandro Flammini

E Ferrara, O Varol, F Menczer, A Flammini.

Traveling trends: social butterflies or frequent fliers?

ACM COSN 2013

Documents

Traveling Trends Social butterflies or frequent fliers?cosn.acm.org/2013/files/Session8/Session8Paper1.pdf · Traveling Trends Social butterflies or frequent fliers? with O. Varol,